Meta places massive order for Graviton 5 cores for AI infrastructure

Meta disclosed the underlying deal weeks before the X posts that circulated on May 19. On April 24, Meta said it had signed an agreement with Amazon Web Services to bring “tens of millions” of AWS Graviton cores into its compute portfolio for agentic AI workloads, making Meta one of the largest Graviton customers in the world. Amazon said the deployment starts with tens of millions of Graviton cores and can expand as Meta’s AI needs grow. Meta said the agreement reflects a “portfolio approach” to infrastructure rather than relying on one chip type for every workload. ### Did Meta actually place a Graviton order, or was that only an X rumor? (about.fb.com) Meta’s own newsroom said on April 24 that it had agreed with AWS to deploy tens of millions of Graviton cores. Amazon’s corporate news site separately described the move as an expansion of a long-standing partnership and said the deployment would begin with tens of millions of cores. (aboutamazon.com) The May 19 X post did not introduce a new transaction date or a new disclosed quantity beyond that. What it did was recirculate an already announced agreement and frame the chips as part of AI orchestration infrastructure. That framing lines up with Meta’s description of Graviton5 as supporting “reasoning, planning, and acting” workloads. (about.fb.com) ### What is Meta saying these chips will do inside its AI stack? Meta said AWS Graviton5 cores are intended for agentic AI workloads rather than the model-training jobs more commonly associated with GPUs. In its April 24 post, Meta said those workloads include “reasoning, planning, and acting,” and said no single chip architecture can efficiently serve every workload. (about.fb.com) Amazon said the agreement builds on Meta’s use of Amazon Bedrock at scale and its broader AWS relationship. The company did not describe Graviton5 as replacing GPUs; instead, both companies presented the CPUs as another layer in Meta’s AI infrastructure mix. ### Why are CPUs part of an AI buildout that usually centers on GPUs? (about.fb.com) AWS has been pitching Graviton5 as a general-purpose processor for workloads that need price-performance rather than the parallel compute profile of a GPU. In its December 2025 launch note for M9g instances, AWS said Graviton5 offers up to 25% better compute performance than Graviton4-based M8g instances, with gains of up to 30% for databases and up to 35% for machine learning workloads. (aboutamazon.com) Meta’s public language suggests those CPUs are being aimed at orchestration and service-layer tasks around AI systems. That is consistent with AWS’s recent Redshift messaging, which says low-latency SQL queries used in analytics pipelines and “autonomous, goal-seeking AI agents” are a target use case for Graviton-powered infrastructure. (aws.amazon.com) ### What does the Redshift RG seminar have to do with this? AWS announced on May 12 that Amazon Redshift RG is a new Graviton-powered instance family for Redshift. AWS said RG instances run data warehouse workloads up to 2.2 times as fast as RA3 instances, data lake queries on Apache Iceberg up to 2.4 times as fast, and come at 30% lower price per vCPU. (aws.amazon.com) The X post’s reference to a June 4 Asia seminar appears tied to that product rollout. AWS’s public materials show Redshift RG as a current launch area and position it for analytics, ETL pipelines, dashboards and agentic AI workloads, which helps explain why Graviton is being discussed alongside AI orchestration rather than only traditional database jobs. (aws.amazon.com) ### How does this fit with Meta’s broader infrastructure push? Meta’s April and May infrastructure posts show the AWS agreement alongside other supplier and data-center moves. Meta’s newsroom lists recent announcements involving NVIDIA, AMD, Broadcom, Arm and new AI-optimized data center projects, indicating the company is spreading AI infrastructure across multiple chip and platform partners. (aws.amazon.com) AWS’s next visible milestone in this thread is June 4, when the company is expected to hold the Asia seminar referenced in the social post on Redshift RG. Meta has not publicly disclosed a dollar value for the Graviton agreement or a deployment completion date beyond saying the rollout starts with tens of millions of cores and can expand. (aboutamazon.com) (about.fb.com)